| The remote sensing image is different from the usual image,it is the original remote sensing data through geometric correction,radiation correction and a series of processes into gray scale and formed the image.Remote sensing technology is entering a new stage that can quickly and accurately provide a wide range of data on earth observation and application research,in the last few decades has been rapid development,and now reach a new climax.Although remote sensing is developing rapidly in both theoretical and application fields,remote sensing is still in transition from qualitative to quantitative,and its accuracy can not fully meet the needs of different users.Especially in the field of remote sensing image detection fields,requires the computer to automatically detect the target,to liberate the manpower from the complex work,the diversity of remote sensing spectral features is also an obstacle to target detection.In this context,this paper takes full account of the diversity of background spectra and target spectra of remote sensing images,using the method of sparse representation to give adaptive weight to the background pixels to suppress,and protect the target pixels,which can increase the difference between the background pixel spectrum and the target pixel spectrum,so that the detector accurately separates the target,the main work is summarized below:1.The method of measuring the similarity between spectra is studied,a method based on sparse representation to measure the similarity of spectral information is proposed,due to the diversity of remote sensing image spectra,only use a single spectrum can not completely characterize the target class,the algorithm constructs a target dictionary that contains only the target spectrum priori information,and use this target dictionary to characterize the target spectral information,then,the detected pixels are sparse represented,finally the restoration residuals are used to measure the similarity between the spectra,this makes the similarity measure more scientific and accurate.2.The difference between increasing background pixel and target pixel is studied,a adaptive sparse weighting method is proposed,by assigning different weights to each pixel in the remote sensing data before the detector begins to detect,to effectively enhance thetarget pixel while suppressing the background pixel,thereby increasing the difference between the background spectral information and the target spectral information,so that the detector can more effectively distinguish the background pixel and the target pixel,thereby improving the detection effect.3.The multi-target detection method of remote sensing image is studied,in this paper,a multi-objective detection method is proposed based on constrained energy minimization algorithm,first,we use the method of sparse weighting to get the spectral characteristic information with large difference,and then the weighted remote sensing data is constrained and optimized by multiple constraints,to achieve the purpose of simultaneously detecting multiple targets. |
